Removed the warnings that are emitted when the LRU caches employed
by the mapper as well as loader srtategies reach their threshold; the
purpose of this warning was at first a guard against excess cache keys
being generated but became basically a check on the "creating many
engines" antipattern.   While this is still an antipattern, the presense
of test suites which both create an engine per test as well as raise
on all warnings will be an inconvenience; it should not be critical
that such test suites change their architecture just for this warning
(though engine-per-test suite is always better).

Change-Id: I41ef8cd642d05a845f53119b196440f9d7879cd9
Fixes: #4071

An :class:`.InvalidRequestError` is raised when a :func:`.synonym`
is used against an attribute that is not against a :class:`.MapperProperty`,
such as an association proxy.  Previously, a recursion overflow would
occur trying to locate non-existent attributes.

Change-Id: If2ce38c429a69951df4c94b71b74edbd59d775e3
Fixes: #4067

Fixed race condition in ORM identity map which would cause objects
to be inappropriately removed during a load operation, causing
duplicate object identities to occur, particularly under joined eager
loading which involves deduplication of objects.  The issue is specific
to garbage collection of weak references and is observed only under the
Pypy interpreter.

Change-Id: I9f6ae3fe5b078f26146af82b15d16f3a549a9032
Fixes: #4068

Refined the behavior of :meth:`.Operators.op` such that in all cases,
if the :paramref:`.Operators.op.is_comparison` flag is set to True,
the return type of the resulting expression will be
:class:`.Boolean`, and if the flag is False, the return type of the
resulting expression will be the same type as that of the left-hand
expression, which is the typical default behavior of other operators.
Also added a new parameter :paramref:`.Operators.op.return_type` as well
as a helper method :meth:`.Operators.bool_op`.

Change-Id: Ifc8553cd4037d741b84b70a9702cbd530f1a9de0
Fixes: #4063

Fixed bug in :meth:`.Session.merge` where objects in a collection that had
the primary key attribute set to ``None`` for a key that is  typically
autoincrementing would be considered to be a database-persisted key for
part of the internal deduplication process, causing only one object to
actually be inserted in the database.

Change-Id: I0a6e00043be0b2979cda33740e1be3b430ecf8c7
Fixes: #4056
(cherry picked from commit 5243341ed886e10a0d3f7fef8ae3d071e0ffdcf0)

Change-Id: Ida0d01ae9bcc0573b86e24fddea620a38c962822

Implement rowcount assertions and single row check
for post_update as well as deletes.

Change-Id: I4e5ba7e8747bf0e0b41f569089eb8cdbf064b7a9
Fixes: #4062

Added a new class of "rowcount support" for dialects that is specific to
when "RETURNING", which on SQL Server looks like "OUTPUT inserted", is in
use, as the PyODBC backend isn't able to give us rowcount on an UPDATE or
DELETE statement when OUTPUT is in effect.  This primarily affects the ORM
when a flush is updating a row that contains server-calcluated values,
raising an error if the backend does not return the expected row count.
PyODBC now states that it supports rowcount except if OUTPUT.inserted is
present, which is taken into account by the ORM during a flush as to
whether it will look for a rowcount.

ORM tests are implicit in existing tests run against PyODBC

Fixes: #4062
Change-Id: Iff17cbe4c7a5742971ed85a4d58660c18cc569c2